Key points are not available for this paper at this time.
This study aims to create a system for recognizing handwritten Devanagari letters using deep neural networks (DNNs) for accurate character identification. The research utilizes explicitly transfer learning techniques in combination with the VGG16 convolutional neural network (CNN) model. The authors have produced a new dataset for this research, which includes around 92,000 images. These images represent 46 characters from the Devanagari script, including consonants and digits. The VGG16 model, previously trained on extensive picture datasets, underwent fine-tuning using our Devanagari character dataset to enhance its performance for this particular application. The model attained an impressive recognition accuracy of 96.58%, a substantial improvement compared to previously reported techniques in the literature. The significant improvement in accuracy demonstrates the effectiveness of the transfer learning method and the resilience of the VGG16 architecture in dealing with the intricacies of handwritten Devanagari letters. The research showcases the promise of sophisticated deep-learning approaches in improving HCR systems. It also thoroughly compares with the previous methodology, highlighting the progress achieved. For future research, the authors intend to investigate deeper learning structures further and integrate a broader and more varied dataset to enhance the model’s accuracy and guarantee its suitability for different real-life situations.
Sharma et al. (Tue,) studied this question.